Increasing taxonomic diversity and spatial resolution clarifies opportunities for protecting US imperiled species.

Continental- and regional-scale assessments of gaps in protected area networks typically use relatively coarse range maps for well documented species groups, creating uncertainty about the fate of unexamined biodiversity and providing insufficient guidance for land managers. By building habitat suitability models for a taxonomically diverse group of 2216 imperiled plants and animals, we revealed comprehensive and detailed protection opportunities in the conterminous United States. Summing protection-weighted range-size rarity (PWRSR, the product of the percent of modeled habitat outside of protected areas and the inverse of modeled habitat extent) uncovered novel patterns of biodiversity importance. Concentrations of unprotected imperiled species in places such as the northern Sierra Nevada, central and northern Arizona, the Rocky Mountains of Utah and Colorado, southeastern Texas, southwestern Arkansas, and Florida's Lake Wales Ridge have rarely if ever been featured in continental- and regional-scale analyses. Inclusion of diverse taxa (vertebrates, freshwater mussels, crayfishes, bumble bees, butterflies, skippers, and vascular plants) partially drove these new patterns. When analyses were restricted to groups typically included in previous studies (birds, mammals, and amphibians), up to 53% of imperiled species in other groups were left out. The finer resolution of modeled inputs (990 m) also resulted in a more geographically dispersed pattern. For example, 90% of the human population of the conterminous United States lives within 50 km of modeled habitat for one or more species with high PWRSR scores. Over one-half of the habitat for 818 species occurs within federally lands managed for biodiversity protection; an additional 360 species have over one-half of their modeled habitat on federal multiple use land. Freshwater animals occur in places with poorer landscape condition but with less exposure to climate change than other groups, suggesting that habitat restoration is an important conservation strategy for these species. The results provide fine-scale, taxonomically diverse inputs for local and regional priority-setting and show that although protection efforts are still widely needed on private lands, notable gains can be achieved by increasing protection status on selected federal lands.

INTRODUCTION

A major goal of conservation is to prevent the decline and extinction of species. This societal value is codified, for example, in multilateral agreements (e.g., the Convention on Biological Diversity's Aichi Targets), national legislation (e.g., the US Endangered Species Act, ESA), and civil society charters. Yet despite more than a century of conservation actions and considerable investment, biodiversity continues to decline (Rosenberg et al., 2019; Tittensor et al., 2014). Site‐level protection (hundreds to thousands of hectares) offers a proven means of preventing species extinction, especially for range‐restricted species (Butchart et al., 2012). Yet even in the United States, where the concept of management of public lands to conserve nature arguably originated, a large proportion of imperiled species occur outside of the current portfolio of protected areas (Jenkins et al., 2015). Without timely and effective conservation efforts, these species will continue their slide toward extinction. In the United States and other countries with endangered species laws, declining species can eventually qualify for legal protection that, while necessary for averting extinction, is often costlier than early prevention (Baruch‐Mordo et al., 2013).

Preventing further declines in imperiled species requires supporting management strategies with accurate information on species distributions. Several efforts have identified general patterns where imperiled species are least protected by existing parks and reserves and therefore in urgent need of conservation (Jenkins et al., 2015; Pimm et al., 2018; Rodrigues et al., 2004). Other analyses demonstrate that range‐restricted species are especially likely to occur outside of protected areas (Akasaka et al., 2016; Gruber et al., 2012; Jenkins et al., 2013; Watson et al., 2011; Wintle et al., 2019).

However, the spatial resolution of most global and many national analyses of gaps in existing protected area networks is limited to the precision of IUCN species range maps (Jenkins et al., 2015; Rodrigues et al., 2004; Watson et al., 2016). These maps are typically the first‐ever electronic depictions of species’ ranges. They provide a valuable perspective on global patterns of diversity and threat but can have limited consistency in approaches for mapping distributions in poorly surveyed areas and large commission errors (i.e., range polygons include areas of unsuitable habitat; Hurlbert & Jetz, 2007; Jetz et al., 2008; McKerrow et al., 2018). They work well to provide insights into continental and regional scale priorities but are less useful for identifying areas for local conservation action.

Researchers continue to improve approaches to estimate species’ distributions to satisfy the needs of conservation planning. For example, combining IUCN range maps with information about habitat preferences and elevational ranges to generate deductive distribution models reduces errors of commission (Rondinini et al., 2011). This method is an improvement over unaltered IUCN range maps but does not address errors and variability in range boundary polygons. Inductive habitat suitability models (HSMs, also known as species distribution models) offer further improvements such as the ability to map distributions at a finer resolution when suitable algorithms, environmental predictors, and valid species occurrence data are used (Sofaer et al., 2019).

Until recently, computational requirements have made finer‐resolution continental‐ or regional‐scale HSMs untenable for analyses of large numbers of species. These models can take advantage of environmental predictor data that are increasingly available at high resolution (e.g., 30 m; Hill et al., 2016, USGS, 2018). Despite the computational burden, the benefits of this information are considerable. For example, with a more comprehensive set of species modeled at a finer resolution, outputs can be compared to land tenure information to formulate more targeted protection strategies. Conserving a concentration of imperiled species on federal lands designated for multiple uses would require a different approach than one on private timber lands. Identifying threats to these multispecies concentrations can be more accurate as well. With sufficient spatial resolution, HSM outputs can provide the foundation for habitat conservation plans, such as those mandated by the US ESA (McAbee et al., 2013).

Availability of species range information limits the taxonomic diversity of global and national analyses of conservation needs. To date, the only globally distributed terrestrial groups with both comprehensive range maps and conservation status available are birds, mammals, and amphibians. Consequently, numerous global assessments are limited to these groups (e.g., Butchart et al., 2012, 2015; Marin & Hedges, 2016; Mason et al., 2020; Moilanen et al., 2013; Tilman et al., 2017; Venter et al., 2014; Watson et al., 2016). Yet these tetrapods do not always perform well as surrogates for other taxonomic groups, especially rare and imperiled species (Darwall et al., 2011; Eglington et al., 2012; Grenyer et al., 2006; Lawler et al., 2003). Invertebrates and plants are typically overlooked in analyses of conservation needs but are important to ecosystem function by providing, for example, habitat, prey, nutrient cycling, water purification, and pollination services (but see Myers et al., 2000). Some of these groups, especially freshwater invertebrates, are typically far more threatened than tetrapods (Collen et al., 2014; Stein et al., 2000).

In addition to taxonomic inputs and spatial resolution, the approach used to identify areas of unprotected imperiled species strongly influences outcomes. Although simple richness of imperiled species may be intuitive to communicate to broad audiences, conservation planners now often focus on either range‐restricted species or range‐size rarity calculations (i.e., the inverse of range size; Guerin & Lowe, 2015; Jenkins et al., 2015) to give greater weight to species with small ranges. These species are more vulnerable to single threatening events and offer fewer opportunities for conservation interventions. Range‐size rarity values summed across analysis units can then be used as inputs into irreplaceability analyses and other site optimization and planning algorithms (Guerin & Lowe, 2015). Further adjustments can account for the degree to which species’ ranges overlap with protected areas to focus on unprotected species (Scott et al., 1993).

Here we present, for the first time, the use of high‐resolution HSMs for a comprehensive set of imperiled species representing diverse taxonomic groups to better identify the areas where unprotected imperiled species occur in the United States. This approach takes advantage of cloud computing, high‐resolution environmental predictor layers, highly vetted species locality data for diverse taxa, and review of preliminary models by taxon experts. Using habitat model outputs for 2216 species, we address the following four questions about the distribution of imperiled species and prospects for conserving them. (1) Where are the concentrations of unprotected imperiled species located? (2) How does increasing spatial resolution and taxonomic diversity influence these patterns? (3) What is the ownership and management status of the lands where imperiled species occur, and how does this influence opportunities for conservation action? (4) Where do poor landscape conditions, high human populations, and projected stresses of climate exposure threaten imperiled species?

The results provide (1) a biodiversity importance layer at a scale suitable for input to local and regional priority‐setting exercises for habitat restoration and land acquisition and (2) insights into where to apply a diverse portfolio of strategies, such as coordination with public land managers, restoration, land acquisition, or climate resilience enhancement, to most effectively conserve imperiled species. We discuss the findings in relation to current conservation policy and opportunities for preventing species from declining to the point of eligibility for ESA listing.

METHODS

Study area and species selection

Our study area was the conterminous 48 United States (hereafter, CONUS), an area for which both high resolution environmental predictor data and highly vetted locality data for imperiled species are available. We included species with global conservation status ranks of critically imperiled (G1) or imperiled (G2) (Faber‐Langendoen et al., 2012) or were listed as Threatened or Endangered under the ESA at the full species level. We restricted species selection to four groups: vertebrates, vascular plants, select freshwater invertebrates (freshwater mussels and crayfishes), and select pollinators (bumble bees, butterflies, and skippers). From the initial list of 2280 species meeting these criteria as of 5 September 2018, we eliminated 64 species for which taxonomic or other uncertainties prevented accurate habitat modeling.

Acquiring habitat suitability models

We obtained or developed HSMs for each species by one of the following methods: (1) procuring existing, vetted, range‐wide models (63 species), (2) developing new HSMs using inductive modeling (1923 species), or (3) for species not amenable to inductive modeling, developing deductive models (230 species). We describe our methods for inductive HSMs below and include methods for deductive models in Appendix S1. We converted data from all sources to binary footprints of suitable habitat/non‐habitat. To smooth out differences caused by using different approaches and different resolutions, we converted habitat models to 990‐m resolution rasters for analysis by selecting all 990‐m cells containing any areas mapped as suitable habitat for a species.

Inductive habitat suitability modeling

We generated HSMs using the random forests algorithm run in R (Breiman, 2001; Liaw & Wiener, 2002) under two frameworks: (1) a raster‐based system for modeling species occurring mostly in terrestrial and palustrine landscapes and (2) a vector‐based system for modeling species occurring in riverine systems. We modeled terrestrial and palustrine species at 30‐m resolution for species with small ranges (less than 100,000 km2) and at 330‐m resolution for species with larger ranges, before resampling to 990‐m resolution for analysis. For riverine species, we used the 2.7 million vector flowlines from the medium resolution National Hydrography Dataset (NHD; EPA, 2012) as the modeling base. We assigned predicted probabilities to both the NHD flowlines and the linked NHD polygons to ensure that larger rivers and lakes were included in the mapped distributions of freshwater species.

We developed all model scripts in R (Microsoft and R Core Team, 2018, with repository snapshot 2019‐02‐01, R Core Team, 2018; available: Data S1, Data S2, and data available online [terrestrial models, Howard et al., 2020b, and aquatic models, Howard et al., 2020a]). We used range information from published sources (IUCN, 2014 [reptiles and amphibians], Kartesz, 2015 [plants], IUCN, 2016 [mammals], BirdLife International and Handbook of the Birds of the World, 2018 [birds], NatureServe, 2019 [fishes, mussels, crayfishes]) as the modeling area for terrestrial and palustrine species and used successively larger USGS hydrological units intersected with presence data for freshwater species. We used NatureServe's Biodiversity Location Data (BLD) as the primary source of model training data and supplemented these with data from specimen and citizen science portals and an unpublished database (Biodiversity Data Serving Our Nation [BISON], Butterflies and Moths of North America [BAMONA], Integrated Digitized Biocollections [iDigBio], iNaturalist, Symbiota Collections of Arthropods Network [SCAN]; L. Richardson, unpublished bumble bee locality data). The BLD provided by state natural heritage programs have all undergone review by local experts before entering databases that typically are used in regulatory review and other highly scrutinized biodiversity applications (Young et al., 2019). We compiled 106 terrestrial and palustrine environmental predictor layers including representations of topography, soils, land cover, and climate, and 172 freshwater variables primarily derived from EPA's StreamCat (Hill et al., 2016) and mapped to NHD flowlines.

For each species, we converted the continuous random forest prediction to a binary representation of habitat suitability by applying a statistically defined probability threshold above which all values were classified as suitable habitat (see Appendix S1). We then uploaded these habitat models to an online model review tool where 86 experts from state natural heritage programs familiar with the species reviewed the outputs (1665 reviews of 1349 species). Reviewers rated the binary prediction on a five‐point scale according to their knowledge of the species (where 1 is completely inconsistent with expected distribution and 5 is good representation of likely habitat) and commented on spatial over‐ or underprediction. We reviewed the models for 685 species that experts were not able to review, focusing on those with poor performance measures and/or limited training data. We did not review models for 148 species with good performance measures and ample training data. Depending on the review results, we (1) used the habitat model as is, (2) modified it by adjusting the threshold or by clipping areas unlikely to support the species, or (3) replaced it with a deductive or alternative model (see Appendix S3: Table S1 for details). The metadata describing the relative importance of predictor variables and model performance for each species is posted in FigShare (Smyth et al., 2020).

Protection‐weighted range‐size rarity and areas of unprotected biodiversity importance

We defined protected areas as those with GAP status 1 and 2 in the Protected Areas Database‐US (PAD‐US; USGS GAP, 2018), combining fee, designation, and easements layers. These areas have a mandate for biodiversity protection and are analogous to IUCN management categories I–IV, which are typically used globally to define protected areas (Jenkins & Joppa, 2009). We made the simplifying assumption that imperiled species occurring within protected areas are effectively protected, and we recognize that some protected areas and conservation easements have not yet been mapped in the PAD‐US.

We calculated protection‐weighted range‐size rarity (PWRSR) for each species as the product of range‐size rarity (the inverse of the modeled habitat area at 990‐m resolution) and the percentage of modeled habitat outside of protected areas, using the habitat data at its native resolution (i.e., 30‐, 330‐, or 990‐m resolution raster or NHD‐polylines and polygons) to more accurately reflect habitat overlap with protected areas than was possible with the 990‐m data.

Using the 990‐m resolution habitat maps clipped to exclude protected areas, we created raster mosaics in ArcGIS to sum PWRSR for all species as well as each of the four taxonomic subsets (vascular plants, vertebrates, freshwater invertebrates, and pollinators). We delineated areas of unprotected biodiversity importance (AUBIs) for use in further analyses by selecting all pixels where summed PWRSR ≥0.0005, an inclusive value designed to highlight areas of conservation value (See Appendix S2 for comparison to other threshold values). A PWRSR score of 0.0005 corresponds to a single species with a range of 1000 km2 that is 50% unprotected, a single species with a range of 20 km2 that is 1% unprotected, or multiple co‐occurring species with lower PWRSR scores.

As a measure of the dispersion of AUBIs relative to human populations, we calculated the portion of the population that lives within 50 km of an AUBI. To do this, we buffered the AUBI layer by 50 km, summed the population count within the buffer (using the Gridded Population of the World data set; CEISIN, 2017), and then calculated the percentage of the CONUS human population within that buffer.

Effect of taxonomic diversity

We explored taxonomic representation in AUBIs by comparing the number of species in each taxonomic group that would be included in a set of AUBIs defined using only amphibians, birds, and mammals (i.e., those typically used). To ensure differences in representation were not caused by the threshold applied, we reran the analysis using PWRSR thresholds from 0.000001 to 0.01 in 20 increments. To ensure that representation was not influenced by the numbers of species in each taxonomic group, we conducted a resampling procedure by randomly sampling all groups with more species than the sum of amphibians, birds, and mammals (103 species) down to 103 species (without replacement) 1000 times. We report the mean and standard deviation of the number of species for the resampled groups (freshwater and anadromous fishes, freshwater invertebrates, plants) and the total numbers of species for the other groups (reptiles, pollinating invertebrates) included in the AUBIs defined using amphibians, birds, and mammals.

Effect of spatial resolution

To investigate the effect of the spatial resolution of the species range inputs on our results, we repeated our analysis for vertebrates using IUCN range maps in place of the HSMs. We assembled range maps for all vertebrates included in our analyses (BirdLife International and Handbook of the Birds of the World, 2018; IUCN, 2014, 2016; NatureServe, 2019), clipped them to the extent of the CONUS to match the extent of the HSM inputs, and repeated the calculation of summed PWRSR. We then compared areas with the highest 5% of summed PWRSR between our original results built on HSMs (vertebrates only) and those mapped using the range maps as inputs.

Stewardship responsibility

We evaluated stewardship responsibility for areas defined as AUBIs as well as the modeled habitat for each species by using the manager type attribute of PAD‐US 2.0 (USGS GAP 2018) and calculating the proportion of modeled habitat overlapping lands in each of three management classes: (1) owned or managed by federal agencies, (2) owned or managed by state or local agencies (includes county, city, and private owners, including non‐profit organizations), or (3) that did not overlap with any PAD‐US layer and is presumed private and unprotected. Federal agency land was further categorized by major federal land managers (Bureau of Land Management, BLM; US Forest Service, USFS; National Park Service, NPS; Fish & Wildlife Service, FWS) using the manager name attribute of the PAD‐US. We calculated both the total area of mapped habitat for each species within each management class, and the total area for each species within the management class but outside protected areas (i.e., areas tagged GAP Statuses 1 and 2). For the BLM and USFS, we repeated the calculation for areas classified as multiple use, or GAP Status 3. We calculated the number of species in each taxonomic group with (1) >50% of their habitat falling within a particular management category (termed “responsibility” species for each land manager) and (2) >90% of their habitat falling within a category (termed ‘endemic’ species), reporting results separately for ESA listed species and for non‐ESA listed species. By definition, AUBIs cannot occur in land tagged GAP Status 1 or 2 and therefore we restricted these latter analyses to species.

Threats

To understand the influence of recognized threats to imperiled species, we calculated average landscape ecological condition and observed trends in climate change exposure across the modeled habitat of each species. For landscape condition, we used a national data layer that reflects the ecological condition of a given place, incorporating infrastructure, land use, and modified vegetation, as well as field validation (Hak & Comer, 2017). Landscape condition is scored on a 0–1 scale, where 0 represents the most altered and 1 represents the most intact landscapes (Appendix S3: Figure S1).

To measure climate change exposure, we used climate “typicality,” which quantifies the departure in recent observed climate from climate variability in a historical baseline (Comer et al., 2019). This metric is a nonparametric, multivariate analog to the standard Euclidean distance metric that has been used in similar contexts (Williams et al., 2007) and has the advantages of not being contingent on the climate data fitting a statistical model and allowing comparison between variables such as temperature and precipitation. To calculate typicality, we used gridded climate data (maximum temperature of the warmest quarter, minimum temperature of the coldest quarter, total precipitation of the wettest quarter, total precipitation of the driest quarter; O'Donnell & Ignizio, 2012) interpolated from weather station measurements obtained from TopoWx (Oyler et al., 2014) and PRISM (Daly et al., 2008). For each climate variable, typicality compares the 30‐yr mean from a recent period (1981–2014) to the full range of values in a baseline period (1948–1980). Specifically, typicality is the proportion of annual values in the baseline period that are greater than the difference between the baseline and current means. The result, for a given pixel, is a value between 0 (high exposure) and 1 (low exposure). The overall typicality value for a pixel is the mean of the univariate typicality measures.

We resampled the landscape condition (990‐m resolution) and climate typicality layers (935‐m resolution) to match the species grids using nearest neighbor interpolation and then calculated the mean value for each species. To test whether taxonomic groups differed in their exposure to these stressors, we used a nonparametric inference test for multivariate data (R package npmv v. 2.4.0; Burchett et al., 2017) that provides both a global test and by‐factor tests without requiring the assumptions of the classical MANOVA. The taxonomic groups were vascular plants, terrestrial vertebrates (birds, mammals, reptiles, amphibians), freshwater animals (fishes, mussels, crayfishes), and pollinators (bumble bees, butterflies, and skippers). We used Wilks’ lambda for both the global and between groups tests with the F approximation and permutation method, respectively. Tests between groups used a closed multiple testing procedure that controlled for type 1 error with a probability threshold of 0.0001 (Burchett et al., 2017).

RESULTS

Modeling results

Species richness and range‐size rarity maps are provided in Appendix S3: Figures S2–S11. Species richness maps, range‐size rarity maps, and protection‐weighted range‐size rarity maps are available at the ArcGIS Living Atlas of the World (ESRI, 2022) under the title of Map of Biodiversity Importance.

Protectedness

Overlay of habitat suitability models and protected areas revealed that the modeled suitable habitat for 295 imperiled species (13.3%) falls completely outside of protected areas. Of these, 223 were plants, 24 were vertebrates, 47 were freshwater invertebrates, and 1 was a pollinator. One thousand twenty‐nine imperiled species (46.4%) have at least 10% of their distributions in protected areas but just 225 of these (10.2% of the total) occur mostly or entirely (>80% of predicted distribution) within protected areas. Taxonomic groups varied in the percentage of suitable habitat that overlapped protected areas (Figure 1).

FIGURE 1

Protectedness (proportion of predicted habitat overlapping protected areas) of imperiled species in the conterminous United States (CONUS; means and SE). Protectedness varied among taxonomic groups (Kruskal‐Wallis χ2 = 54.52, df = 2, p < 0.0001; Dunn's post hoc tests identified significantly [p ≤ 0.05] different groups, which are marked by different letters; numbers are sample sizes). The mid‐line indicates the median, the box edges indicate the 25th and 75th percentile (interquartile range) of the data, and whiskers indicate points within 1.5 times the interquartile range. The points indicate outliers beyond the whisker range

Concentrations of protection‐weighted range‐size rarity of imperiled species

Concentrations of protection‐weighted range‐size rarity of imperiled species occurred in California, scattered areas across the West and Southwest, Texas, Arkansas, and the Southeast (Figure 2). AUBIs covered 510,521.3 km2 (6.3%) of the CONUS (Figure 2b). The AUBIs tended to identify areas of multiple imperiled species: 86% of AUBIs had more than one imperiled species modeled to occur in them, with a mean of 4.8 species (SD 3.21) predicted to occur in each. Overall, the habitat of 2124 species (96% of all modeled species) was predicted to occur in an AUBI.

FIGURE 2

Protection‐weighted range‐size rarity (PWRSR) of imperiled species in the conterminous United States. Upper panel shows actual values. Lower panel shows areas of biodiversity importance (all pixels with values >0.0005)

By far, California had the greatest AUBI area, followed at some distance by Texas and then states in the Southeast and West (Table 1, Appendix S2: Table S1). California also had the greatest percentage of its land area in AUBIs (27.6%) (Appendix S2: Table S1). Florida, Georgia, and Tennessee also had nearly a fifth of their land area overlapped by AUBIs. AUBIs occurred in all states except Rhode Island. Ninety percent of the human population of the CONUS lives within 50 km of an AUBI.

TABLE 1

Top 10 states with the greatest extent of areas of unprotected biodiversity importance defined for all taxa, plants, vertebrates, freshwater invertebrates, and pollinators

State	Area identified as AUBIs (km2)	Coverage of state (%)
All taxa
California	113,026	27.6
Texas	42,100	6.1
Arizona	30,077	10.2
Florida	29,778	20.3
Georgia	29,757	19.5
Utah	28,606	13.0
Colorado	24,901	9.2
Alabama	22,997	17.2
Tennessee	21,070	19.3
Oregon	18,252	7.3
Plants
California	99,560	24.0
Colorado	24,501	9.0
Utah	24,173	11.0
Arizona	23,037	8.0
Texas	20,712	3.0
Florida	19,871	14.0
Oregon	14,070	6.0
New Mexico	13,377	4.0
Georgia	12,816	8.0
Wyoming	12,145	5.0
Vertebrates
Tennessee	12,652	12.0
California	11,433	3.0
Texas	10,038	1.0
Georgia	9152	6.0
Alabama	8940	7.0
Mississippi	5790	5.0
North Carolina	4849	4.0
Virginia	4774	5.0
Florida	4228	3.0
Arizona	3806	1.0
Freshwater invertebrates
Georgia	10,111	6.6
Tennessee	7972	7.3
Alabama	7597	5.7
Texas	5902	0.9
North Carolina	5459	4.3
Florida	5218	3.6
Arkansas	4725	3.4
Kentucky	3464	3.3
Virginia	2939	2.8
Mississippi	2119	1.7
Pollinators
California	1325	0.3
Oregon	1029	0.4
New Mexico	1000	0.3
Arizona	523	0.2
Louisiana	216	0.2
Mississippi	192	0.2
Texas	181	0.0
Nevada	62	0.0
North Carolina	11	0.0

Note: Only nine states are listed for pollinators because those were the only states that had AUBIs defined for this group.

AUBIs defined only by plants (i.e., the summed protection‐weighted range‐size rarity of imperiled plants in a pixel was >0.0005) occurred throughout the CONUS with particular concentrations in the West and in Texas (Figure 3). Vertebrate AUBIs were concentrated in the Southeast with additional AUBIs in the West where many imperiled endemic fishes occur (Figure 3). Freshwater invertebrate AUBIs were centered in the Southeast, whereas pollinator AUBIs were broadly distributed across the southern United States and north into Oregon (Figure 3). California had the greatest area of plant and pollinator AUBIs whereas Georgia had the greatest area of freshwater invertebrate AUBIs and Tennessee had the greatest area of vertebrate AUBIs (Table 1). Texas is noteworthy in being the only state included in the 10 states with the greatest AUBI area for each of the four taxonomic groupings (Table 1).

FIGURE 3

Protection‐weighted range‐size rarity (PWRSR) of imperiled species by taxonomic group: (a) vascular pants, (b) vertebrates, (c) freshwater invertebrates (freshwater mussels and crayfishes), (d) pollinators (butterflies, skippers, and bumble bees)

Effect of distribution map spatial resolution

Use of habitat models to calculate PWRSR of vertebrates gives a more nuanced view of opportunities for protecting imperiled vertebrates compared to use of range maps (Figure 4). Range map input identified 850 unique polygons with high PWRSR and a mean size of 467 km2 (SD 4541) while the higher resolution habitat models identified 11,919 much smaller polygons (mean 33 km2, SD 384). Although the southern Appalachians, Florida Panhandle, parts of Texas, and California's Central Valley and southern mountains are highlighted as important using either type of input maps (28% of the areas overlap), the habitat models also point to many other important areas (Figure 4). The vertebrate habitat models identify important areas in 45 of the 48 states, while the range maps do so in only 24 states (Appendix S3: Table S2).

FIGURE 4

Comparison of using range maps (low resolution, likely high rate of commission error) versus habitat suitability models (high resolution, lower rate of commission error) in identifying areas of high summed protection‐weighted range‐size rarity (PWRSR) for imperiled vertebrates. The highest 5% of values within CONUS are shown for each type of distribution map

Using only birds, mammals, and amphibians to define AUBIs resulted in large numbers of other imperiled species being unrepresented in the AUBIs (Figure 5). For example, for a PWRSR threshold of 0.0000183, the largest threshold that still includes 99% of the modeled ranges of imperiled birds, mammals, and amphibians, 53% of imperiled freshwater invertebrates, 51% of imperiled fishes, and 42% of imperiled plants would be excluded.

FIGURE 5

Percentage of species for which modeled distributions overlap of areas of unprotected biodiversity importance (AUBIs) defined by birds, mammals, and amphibians at different thresholds of protection‐weighted range‐size rarity (PWRSR). For freshwater invertebrates, freshwater and anadromous fishes, and plants, which have more imperiled species than the birds/mammals/amphibians group (103 species), the values shown are the means and standard deviations of random resampling down to the same number of species as the birds/mammals/amphibians group

AUBIs were primarily on private land (65.9%), followed by federal (26.9%), state (4.6%), and local government jurisdictions (2.8%). For a slight majority (52.2%) of species, over one‐half of the modeled habitat occurs on private lands outside of management by federal, state, local, or a combination of these governmental agencies (Table 2). Federal agencies manage land for 818 (36.9% of the species studied) responsibility species (i.e., >50% of habitat is in agencies’ land), 711 (86.9% of those on federal lands) of which are not ESA protected (Table 2).

TABLE 2

Management responsibility for US imperiled species

Management authority for >50% of modeled habitat	Plants		Vertebrates		Freshwater invertebrates		Pollinators		Totals
	ESA	Other imperiled	ESA	Other imperiled	ESA	Other imperiled	ESA	Other imperiled	ESA	Other imperiled	All
Federal Agency
BLM	23	170	6	5	0	0	0	1	29	176	205
USFS	20	346	15	18	3	9	0	7	38	380	418
NPS	12	71	3	6	0	0	0	3	15	80	95
FWS	2	13	2	0	0	0	0	0	4	13	17
Other or Mixed a	10	51	11	8	0	1	0	2	21	62	83
Federal total	67	651	37	37	3	10	0	13	107	711	818
State & local	39	72	8	10	2	4	0	1	49	87	136
Private b	199	529	92	107	69	135	3	22	363	793	1156
Mixed c	22	62	10	9	3	2	0	4	35	77	112
Total d	324	1312	146	163	77	151	3	40	550	1666	2216

Note: Shown are numbers of species for which over half of their modeled habitat extent is on land with different management authorities. Federal agency abbreviations: BLM, Bureau of Land Management; FWS, Fish & Wildlife Service; NPS, National Park Service; USFS, US Forest Service. ESA: Species listed as threatened or endangered under the US Endangered Species Act. Other imperiled: Species categorized as critically imperiled (G1) or imperiled (G2) by NatureServe, but that are not ESA listed.

Species for which more than half of their modeled habitat is on land managed by another federal agency (such as the Department of Defense or Bureau of Land Reclamation) or for which the sum of modeled habitat managed across multiple federal agencies is greater than 50% of the overall modeled distribution.

Private lands are those with no ownership category in the PAD‐US 2.0.

Species for which no single manager class is responsible for more than 50% of the modeled distribution.

The total is slightly less than the sum of federal, state and local, presumed private, and mixed because some species occur on land that is jointly managed by states and the federal government, and those species are counted in both state and federal categories.

The federal responsibility species were disproportionately plants, which make up 73.8% of the species studied but 87.8% of federal responsibility species (Fisher's Exact Test p < 0.0001). Pollinator responsibility species occur on federal land in a proportion similar to their fraction of the study species (1.9% of study species and 1.6% of responsibility species on federal land, p = 0.64), whereas freshwater invertebrate (10% of study species but 1.6% of responsibility species on federal land, p < 0.0001) and vertebrate (14% of study species, 8% of responsibility species on federal land, p < 0.001) responsibility species are underrepresented on federally managed lands.

Three hundred eighty‐two of the species were federal endemics (i.e., only occurring within federal lands; Appendix S3: Table S3). Of these, only 28 species are ESA listed and the remaining 354 imperiled species do not currently receive legal protection. States manage land for 36 endemics (including nine that are ESA listed), and 443 species (including 122 ESA listed species) are endemic to private lands.

Multiple‐use (GAP Status 3) BLM and USFS lands host 360 of the federal responsibility species. This figure includes 70 species endemic to GAP Status 3 lands. The USFS manages land where most of these species occur (245 responsibility species and 57 endemics; Appendix S2: Table S4a), and the BLM manages the remaining 115 responsibility and 13 endemic species (Appendix S2: Table S4b). Most of these species on BLM and USFS GAP Status 3 land are not protected by the ESA: 326 of the responsibility species are imperiled but not ESA listed, as are 64 of the endemics.

Landscape condition and climate exposure

Climate Typicality values ranged from 0.321 to 0.897 and Landscape Condition scores ranged from 0 to 100 averaged across each species’ modeled distribution. (Figure 6). The global inference test indicated differences among taxonomic groups (Wilk's lambda; test statistic = 77.166, df1 = 6.00, df2 = 4410.00, p < 0.0001) and the permutation tests indicated that freshwater animals had significantly higher typicality scores (p < 0.0001) and lower condition scores (p < 0.0001) than the other groups, individually and combined. There were no significant differences in scores for any of the other groups.

FIGURE 6

Landscape condition and observed climate typicality (a climate exposure metric) of modeled habitats of imperiled freshwater animals (fishes, mussels, crayfishes), pollinators (bumble bees, butterflies, skippers), terrestrial vertebrates, and vascular plants. Each curve depicts the density of species for (a) different values of landscape condition (100 = good condition) and (b) climate typicality (1.0 = typical climate or low exposure); the asterisk indicates that the density for freshwater animals differed from the other groups (Wilks’ lambda, p < 0.0001)

DISCUSSION

Leveraging high resolution environmental predictor layers, highly vetted occurrence data, cloud processing power, and dozens of field biologists familiar with the species, we successfully compiled modeled distributions for 2216 imperiled species in the CONUS. Combined with recently updated protected area boundaries, we confirmed previous observations that most imperiled fauna and flora fall outside of protected areas (Jenkins et al., 2015) and provide a novel map of areas with concentrations of unprotected imperiled species at a scale (990 m) appropriate for local and regional conservation priority setting. Unlike previous efforts that scale down from range maps, these results scale up (in most cases) from 30‐m resolution habitat suitability models, resulting in more refined predictions of where target species are likely to occur. These analyses provide a fresh context for understanding conservation needs and opportunities, as well as threats to imperiled species across the CONUS.

The map of AUBIs identifies where site‐ and regional‐scale land and water protection might be particularly effective at preventing species from declining toward extinction. The output offers a powerful input for priority‐setting exercises that additionally account for the conservation values of the entities that will act on the outcomes (e.g., birds for a bird‐focused organization or plants for a state native plant society), spatial distributions of costs, distributions of “coarse filter” vegetation communities, political viability (e.g., potential support or opposition of local communities), and complementarity of sites, among other considerations (Brown et al., 2015). The outputs subsetted by taxonomic groups or ecological guilds can similarly support more targeted priority setting. For example, an exercise to identify priorities for protecting biodiversity in a river drainage might include models for only the freshwater invertebrates. In addition, researchers can use the continuous PWRSR output and select a more appropriate threshold for the taxa of interest.

Geography of unprotected imperiled species

High concentrations of unprotected, range‐restricted imperiled species occurred at numerous locations across the southern and western portions of the country, with lower concentrations dispersed more broadly. These concentrations are perhaps most remarkable for their extensiveness and distribution across multiple regions of the country. Despite the restrictive threshold we used to define AUBIs, they nonetheless cover over 6% of the CONUS land area and occur in 47 of the 48 states.

Our finding that 90% of the CONUS population lives within 50 km of an AUBI demonstrates that both opportunities and the responsibility to protect globally imperiled species are widely distributed. Human populations near AUBIs may increase threats due to encroachment in habitats utilized by imperiled species. Alternatively, this proximity could facilitate education about the co‐benefits and ecosystem services (carbon sequestration, water quality, air quality, recreation, viewshed, public health) obtained from protecting habitat for rare species (Chan et al., 2011).

A previous analysis of protection needs in the United States using vertebrates and trees identified nine priority areas: (1) middle to southern Blue Ridge Mountains; (2) southern Sierra Nevada; (3) California Coast Ranges; (4) Tennessee, Alabama, and northern Georgia; (5) Florida panhandle; (6) Florida Keys; (7) Klamath Mountains in Oregon and California; (8) south‐central Texas; and (9) California Channel Islands (Jenkins et al., 2015). We confirm the occurrence of concentrations of unprotected, range‐restricted imperiled species at all of these areas except the California Channel Islands, which are mostly protected. However, we also find large concentrations of unprotected, range‐restricted, imperiled species in the northern Sierra Nevada, central and northern Arizona, the Rocky Mountains of Utah and Colorado, southeastern Texas, southwestern Arkansas, and the Lake Wales Ridge in south‐central Florida. Range‐restricted, imperiled plants contribute the majority of the PWRSR scores for the AUBIs in many of these previously unidentified areas.

At the state level, California stands out as having numerous opportunities to conserve currently unprotected, range‐restricted imperiled species. With AUBIs covering over one‐quarter of the land area (and protected areas another 27%), much of the state either harbors or is in close proximity to suitable habitat for these species. Although AUBIs cover just 6.1% of Texas, the state nonetheless stands out for its diversity of habitats that place it as the third to seventh most important state in terms of AUBI area for the four taxonomic groups addressed in the study. Conservation efforts in Texas therefore present opportunities to protect a wide variety of imperiled taxa.

One mechanism that every US state has for protecting their biodiversity is a State Wildlife Action Plan (SWAP; federally mandated plans that describe state actions to conserve wildlife). SWAPs can provide direction for funding land and water protection for imperiled species, often aimed at preventing further declines to avoid the need for ESA listing. Unfortunately, most SWAPs do not include plants among their species of greatest conservation need (Stein & Gravuer, 2008; USGS, 2019), despite the large numbers of imperiled plants that fall outside of protected areas.

Use of finer spatial resolution in input maps

Although both the range map and habitat model analyses started from the same list of vertebrate species, the results differed substantially in the geographic spread, number, and size of the highest‐scoring areas of PWRSR. The larger areas of individual species range maps as compared to habitat models contributed to these differences. For example, range maps for freshwater fishes are delineated by the watersheds where each species occurred, whereas the habitat models used only river segments and waterbodies. Consequently, range‐size rarity scores differed, as did calculation of percent of ranges protected.

Use of habitat models derived from observations facilitated the detection of small, discrete areas of importance for biodiversity that would be overlooked with the range map approach. Although conserving large land tracts is important to maintain populations of animals with large home ranges and to reduce edge effects and fragmentation (Trombulak & Baldwin, 2010), small tracts can be important for complementarity and representativeness of protected area networks (Wintle et al., 2019) as well as for protection of smaller animals (Longcore & Osborne, 2015) and rare plants (Akasaka et al., 2016). Indeed, the PWRSR results for the broader suite of plant and animal species analyzed emphasizes the need for protection of many relatively small areas to more fully conserve the nation's biodiversity.

Use of greater taxonomic diversity

Our finding that the distribution of unprotected imperiled diversity varied by taxonomic group was no surprise. Decades of research on surrogacy for conservation planning has demonstrated that taxonomic groups rarely display large degrees of spatial correlation in measures of richness of all species, restricted‐range species, or threatened species (Rodrigues & Brooks, 2007; Ward et al., 2020; Yong et al., 2018). This is true both in the geography of where species occur (Rodrigues & Brooks, 2007; Ward et al., 2020) and in the spatial results of optimized reserve site selection exercises (Yong et al., 2018). One possible exception to this rule is that species diversity may serve as a surrogate for phylogenetic and functional diversity in New World tetrapods (Rapacciuolo et al., 2018).

In the United States, AUBIs identified with data only from birds, mammals, and amphibians, the most widely used combination of taxonomic groups to assess biodiversity and conservation patterns globally, had limited overlap with the modeled ranges of other taxonomic groups. Aside from a geographical mismatch, one reason for the lack of overlap among taxonomic groups is the limited number of imperiled birds, mammals, and amphibians (103 species). Birds and mammals are among the taxa with the lowest levels of imperilment in the United States (Stein et al., 2000). Amphibians, like other freshwater species, have much higher levels of imperilment but their small distributions collectively also were responsible for a small percentage of the AUBI area. Even controlling for numbers of imperiled species as our resampling of the more speciose groups accomplished, modeled ranges of large numbers of imperiled species did not overlap with the bird/mammal/amphibian AUBIs due to geographically distinct distributions.

We found that unprotected imperiled plants are especially noteworthy for their discordant distribution from much of the area where unprotected imperiled animals are located. At least in the CONUS, many more plants are considered imperiled than animals (1636 plants vs. 580 animals of the groups examined in this study). With increasing availability of data on plant distribution and imperilment status, researchers can now more easily consider this important component of the US's imperiled biodiversity. As a first step, more states should include plants in their SWAPs where they will receive greater attention for conservation actions.

Even the extensive taxonomic breadth of species used in this study is just a subset of the known imperiled terrestrial and freshwater species in the United States. Groups that have been comprehensively assessed for conservation status in the CONUS by NatureServe and/or the IUCN include terrestrial snails, several freshwater animal taxa (freshwater snails; fairy, clam, and tadpole shrimps; stoneflies; mayflies; caddisflies; dragonflies; and damselflies), other insects (several macro moth and bee groups, tiger beetles), nonvascular plants, and lichens. Although we now know which of these species qualify as imperiled, range‐wide locality data required for modeling habitat suitability are often not available. Not all state natural heritage programs have either the mandate or the capacity to compile geospatial data on the locations of populations, and digitization of natural history museum specimen data is only just beginning for many of these groups. While the rapidly increasing availability of locality data from citizen science programs can yield robust data sets for species identifiable through images and other digital media, many rare species require continued investment in professional field surveys (Young et al., 2019).

Established protected areas clearly have a major role to play in protecting US biodiversity: over one‐half of the imperiled species studied are likely to occur inside protected areas. Nevertheless, large numbers of imperiled species remain outside of the current conservation estate. More species have the majority of their modeled habitats on private land than in any other stewardship class, confirming an observation that has been made previously (Jenkins et al., 2015). The conservation community, federal agencies, and private landowners have developed numerous strategies for protecting species on private lands, including private reserves, easements, tax incentives, Candidate Conservation Agreements, environmental and land use regulation, landowner‐directed habitat enhancement incentives, and joint ventures (Kamal et al., 2013; Schuster et al., 2018; Young et al., 2019). These approaches, together with creative new ones, can be informed by the spatial precision of outputs presented here to help recover populations of imperiled species.

Yet the finding that 326 imperiled, non‐ESA listed species occur primarily on federal multiple‐use lands highlights an essential responsibility of federal agencies to balance conservation needs with other land uses. With federal management authority already in place, some of the obstacles inherent to conservation on private lands, such as land acquisition costs, are eliminated. Both the USFS and BLM have policies and programs that can steer incompatible uses away from populations of imperiled species. Examples include the USFS Forest Planning Rule and BLM Areas of Critical Environmental Concern (ACEC) program. The detailed outputs of this study can help inform site and project‐specific planning to reduce threats to imperiled species on federal lands. We emphasize in addition to re‐designating some multiple use land to more restrictive GAP Status 1 and 2 categories, public land agency managers can use our results combined with field surveys to precisely map compatible use zones in their planning process.

Strategic protective designation of federal lands may nevertheless be the best approach for preventing declines in some species. For example, stricter protection of a 100‐km2 area of Lincoln National Forest (New Mexico) could result in a 25%–74% increase in protection for each of four plant species. If field surveys confirm the presence of viable populations of these imperiled species, reclassifying this relatively small portion of the Forest for biodiversity protection could provide a high return on conservation investment. Another example is an area in the degazetted portion of Grand Staircase‐Escalante National Monument owned by BLM, which was removed from protected status in 2018. This area contains modeled habitat for one imperiled bird and six imperiled plants in our study, including over 95% of the modeled habitat for Xylorhiza confertifolia and Penstemon ammophilus. Designation of selected parts of this area to focus more on biodiversity protection (e.g., assigning ACEC status, or re‐gazetting to national monument status) could substantially improve the outlook for these species.

Nearly one‐third of the imperiled species included in this study are plants that occur primarily on federal lands. Federal agencies, therefore, undisputedly have a major responsibility for plant conservation in the US. Although some states have their own plant conservation strategies (e.g., New Mexico; Roth, 2017), the United States surprisingly does not have a national plant conservation strategy apart from an early continental effort (Maina & Villa‐Lobos, 1997). The spatial data outputs from this study can provide the foundation for a national plant conservation strategy. Until such an effort is undertaken, our results represent an important advancement in characterizing the geography of the nation's imperiled vascular plants.

Although this study and others (Dickson et al., 2014; Scott et al., 1993; Stein et al., 2008) demonstrate the importance and potential of federal lands for biodiversity conservation, most imperiled vertebrates and freshwater invertebrates occur primarily on private lands. Imperiled freshwater fishes and invertebrates in the United States are concentrated in the Southeast, whereas federally managed protected areas are most extensive in the West. Conservation of freshwater biodiversity often involves many other strategies besides land protection, including restoration of flow dynamics, protection of riparian ecosystems, dam management and removal, and reduction of contaminants. This study identifies the freshwater systems where such actions would have a high impact on imperiled species having particularly narrow ranges.

Analysis of landscape condition and climate exposure further emphasizes the exceptional case of imperiled freshwater animals, which, on average, occur in areas with higher direct anthropogenic threats but less observed climate change exposure than species in other groups. Centuries of land use change since European settlement and less extensive protected areas in the Southeast, where most imperiled freshwater species occur, have left a landscape more transformed than almost anywhere else in the CONUS outside of the Midwest and the Great Lakes region (Hak & Comer, 2017). Although the landscape condition analysis is based on terrestrial parameters such as transportation corridors, mines, and agricultural activity (Hak & Comer, 2017), these land uses damage freshwater systems through altered hydrology, siltation, runoff of pesticides and mine tailings, and water extraction and diversion. Assuming the pattern of less altered climates in the Southeast continues, maintaining the hydrological needs for the region's freshwater animals in the near term could be a realistic goal.

Imperiled species in the CONUS occur across the spectrum of landscape condition and climate exposure. Species that occur in areas of good landscape condition and low climate exposure are candidates for traditional conservation strategies to protect land from conversion via whatever mechanism is most practical in the local context. For species inhabiting areas with poor landscape condition but limited climate exposure, such as many imperiled freshwater animals, ecosystem restoration will likely need to be part of the strategy. Where species occur in good landscape condition but high climate exposure, protection should focus on adaptation to the specific vulnerability that species have to climate change (Foden et al., 2018). The greatest challenge is for species that occur in areas of poor landscape condition and high climate exposure. In these situations, conservationists may need to consider more radical interventions such as translocations (Prober et al., 2019; Thomas, 2011).

Modeling approach

We incorporated three methodological improvements compared to most previous biodiversity analyses that used large numbers of habitat suitability models. First, as model input we used highly vetted locality data derived from decades of national standardized biodiversity inventory conducted by the nation's network of Natural Heritage Programs. Second, most models were run at a 30‐m spatial resolution, a scale more typically used for generating models intended to inform management of high‐profile species (Sofaer et al., 2019). This approach took advantage of high‐resolution data for over 250 environmental predictor variables before upscaling to 990‐m resolution for the combined analyses. Third, biologists familiar with the species helped remove areas of overprediction, adjust binary map thresholds, and identify data gaps.

Although the methods improve upon previous modeling studies, the models could be further improved for consequential, species‐specific decisions such as those associated with regulatory mandates or habitat management in areas of oil, gas, mineral, and timber extraction. To increase the quality of and confidence in modeled distributions, more effort to review additional sources of locality data, incorporate true absence data, tailor environmental predictors to species traits, elicit wider expert review, validate with newly collected field survey data, and use of ensemble models would all help to improve model quality (Sofaer et al., 2019). Nevertheless, the models created here serve as valuable precursors to more refined species habitat models. They can also inform field surveys to determine species’ occurrences and absences in understudied areas or where they were known to occur historically.

Caveats

For the purpose of identifying geographic patterns of conservation need, we selected a protection‐weighted range‐size rarity threshold appropriate for the national‐level analysis presented here. We emphasize that the main product of the study is a spatial dataset with a continuous scale of summed protection‐weighted range‐size rarity. Local use of these data layers may require threshold adjustments. For example, places with lower levels of protection‐weighted range‐size rarity, such as Rhode Island or South Dakota, may benefit from lower thresholds than areas with higher levels, such as California or Texas.

Our analyses assumed that protected areas are effectively managed for species that occur in them, and that protected areas protect freshwater species, which could be threatened by upstream activities occurring outside the protected area. This may not always be the case due to climate change, local management priorities, and limitations in management capacity (Watson et al., 2014). Also, additional refinements to the GAP Status definition and ranks are needed to capture the important contributions to regional conservation that state, regional, and local resource and conservation agencies provide. The Status categories are sometimes inconsistently assigned in non‐federal public and private conservation lands, and as such many of the important stewardship and restoration actions that are implemented on these lands are not captured in conservation status assessments.

The modeled habitat data we used are hypotheses and therefore require field validation. Nevertheless, HSMs based on vetted observation and high‐resolution predictor variables are much more spatially precise at directing field surveys than traditional range maps. Range maps are still valuable for countless conservation applications, especially where high quality modeled distributions are not available (e.g., Rapacciuolo et al., 2018).

CONCLUSIONS

This study proves the feasibility and demonstrates the importance of developing high quality habitat models for large numbers of species of diverse taxonomic affinities for conservation planning. Doing so dramatically expands the map for where conservation actions can protect imperiled species, and points to targeted actions in specific geographies. Continuing advances in digital field surveys, near real‐time environmental predictor data, cloud computing, and biodiversity information infrastructure can facilitate the future development of a dynamic, updateable national library of habitat models for all imperiled species. Such a resource would generate significant benefits for conserving species through more effective land and water protection, clearer regulatory decisions, and strategies for mitigating threats such as climate change. Yet the current study provides local and regional conservation planners with taxonomically diverse and geographically refined data to achieve efficient and effective conservation actions today.

CONFLICT OF INTEREST

The authors declare no conflict of interest.

AUTHOR CONTRIBUTIONS

S. Breyer, D. Cameron, and H. Hamilton conceived of the project. D. Cameron, C. Frye, H. Hamilton, T. Howard, C. Schloss, R. Smyth, C. Tracey, and B. Young determined the methods. A. Chazal oversaw compilation of existing models and locality data. A. Conley, T. Howard, R. Smyth, and C. Tracey ran the habitat suitability models. R. Smyth oversaw model review. C. Frye performed spatial analyses involving the PAD‐US. B. Young wrote all drafts of the paper. All authors gave final approval for publication.

Appendix S1

Click here for additional data file.

Appendix S2

Click here for additional data file.

Appendix S3

Click here for additional data file.

Data S1

Click here for additional data file.

Data S2

Click here for additional data file.